Cost-based Query Optimization

1. +
Cost-based Query Optimization
Maryann Xue (Intel)
Julian Hyde (Hortonworks)
Hadoop Summit, San Jose
June 2016

2. •@maryannxue
•Apache Phoenix PMC member
•Intel
•@julianhyde
•Apache Calcite VP
•Hortonworks

3. What is Apache Phoenix?
• A relational database layer for Apache HBase
– Query engine
• Transforms SQL queries into native HBase API calls
• Pushes as much work as possible onto the cluster for parallel
execution
– Metadata repository
• Typed access to data stored in HBase tables
– Transaction support
– Table Statistics
– A JDBC driver

4. Advanced Features
• Secondary indexes
• Strong SQL standard compliance
• Windowed aggregates
• Connectivity (e.g. remote JDBC driver, ODBC driver)
Created architectural pain… We decided to do it right!

5. Example 1: Optimizing Secondary Indexes
How we match secondary
indexes in Phoenix 4.8:
What about both?
SELECT * FROM Emp ORDER BY name
SELECT * FROM Emp WHERE empId > 100
CREATE TABLE Emps(empId INT PRIMARY KEY, name VARCHAR(100));
CREATE INDEX I_Emps_Name ON Emps(name);
SELECT * FROM Emp
WHERE empId > 100 ORDER BY name
Q1
Q2
Q3
I_Emps_Name
Emps
We need to make a cost-based decision! Statistics can help.
?

6. Phoenix + Calcite
• Both are Apache projects
• Involves changes to both projects
• Work is being done on a branch of Phoenix, with changes to Calcite
as needed
• Goals:
– Remove code! (Use Calcite’s SQL parser, validator)
– Improve planning (Faster planning, faster queries)
– Improve SQL compliance
– Some “free” SQL features (e.g. WITH, scalar subquery, FILTER)
– Close to full compatibility with current Phoenix SQL and APIs
• Status: beta, expected GA: late 2016

7. Current Phoenix Architecture
Parser
Algebra
Phoenix Schema
Stage 1: ParseNode tree
Stage 2: Normalization,
secondary index rewrite
Stage 3: Expression tree
HBase Data
Runtime
Query Plan

8. Calcite Architecture
Parser
Algebra
Schema SPI Operators,
Rules,
Statistics,
Cost model
Data
Engine
Data
Engine
Data
Engine

9. Phoenix + Calcite Architecture
Parser
Algebra
Phoenix Schema Logical + Phoenix Operators,
Builtin + Phoenix Rules,
Phoenix Statistics,
Phoenix Cost model
Data
JDBC (optional)
HBase Data
Phoenix Runtime
Data
Other (optional)
Query Plan

10. Cost-based Query Optimizer
with Apache Calcite
• Base all query optimization decisions on cost
– Filter push down; range scan vs. skip scan
– Hash aggregate vs. stream aggregate vs. partial stream aggregate
– Sort optimized out; sort/limit push through; fwd/rev/unordered
scan
– Hash join vs. merge join; join ordering
– Use of data table vs. index table
– All above (any many others) COMBINED
• Query optimizations are modeled as pluggable rules

11. Calcite Algebra
SELECT products.name, COUNT(*)
FROM sales
JOIN products USING (productId)
WHERE sales.discount IS NOT NULL
GROUP BY products.name
ORDER BY COUNT(*) DESC
scan
[products]
scan
[sales]
join
filter
aggregate
sort
translate SQL to
relational
algebra

12. Example 2: FilterIntoJoinRule
SELECT products.name, COUNT(*)
FROM sales
JOIN products USING (productId)
WHERE sales.discount IS NOT NULL
GROUP BY products.name
ORDER BY COUNT(*) DESC
scan
[products]
scan
[sales]
join
filter
aggregate
sort
scan
[products]
scan
[sales]
filter’
join’
aggregate
sort
FilterIntoJoinRule
translate SQL to
relational
algebra

13. Example 3: Phoenix Joins
• Hash join vs. Sort merge join
– Hash join good for: either input is small
– Sort merge join good for: both inputs are big
– Hash join downside: potential OOM
– Sort merge join downside: extra sorting required sometimes
• Better to exploit the sortedness of join input
• Better to exploit the sortedness of join output

14. Example 3: Calcite Algebra
SELECT empid, e.name, d.name, location
FROM emps AS e
JOIN depts AS d USING (deptno)
ORDER BY d.deptno
scan
[emps]
scan
[depts]
join
sort
project
translate SQL to
relational
algebra

15. Example 3: Plan Candidates
scan
[emps]
scan
[depts]
hash-join
sort
project
scan
[emps]
scan
[depts]
sort
merge-join
projectCandidate 1:
hash-join
*also what standalone
Phoenix compiler
would generate.
Candidate 2:
merge-join
1. Very little difference in all other operators: project, scan, hash-join or merge-join
2. Candidate 1 would sort “emps join depts”, while candidate 2 would only sort “emps”
Win
SortRemoveRule
sorted on [deptno]
SortRemoveRule
sorted on [e.deptno]

16. Example 3: Improved Plan
scan ‘depts’
send ‘depts’ over to RS
& build hash-cache
scan ‘emps’ hash-join ‘depts’
sort joined table on ‘e.deptno’
scan ‘emps’
merge-join ‘emps’ and ‘depts’
sort by ‘deptno’
scan ‘depts’
Old vs. New
1. Exploited the sortedness of join input
2. Exploited the sortedness of join output

17. (and now, a brief look at Calcite)

18. Apache Calcite
• Apache top-level project since October, 2015
• Query planning framework
– Relational algebra, rewrite
rules
– Cost model & statistics
– Federation via adapters
– Extensible
• Packaging
– Library
– Optional SQL parser, JDBC server
– Community-authored rules, adapters
Embedded Adapters Streaming
Apache Drill
Apache Hive
Apache Kylin
Apache Phoenix*
Cascading
Lingual
Apache Cassandra*
Apache Spark
CSV
In-memory
JDBC
JSON
MongoDB
Splunk
Web tables
Apache Flink*
Apache Samza
Apache Storm

19. Apache Calcite Avatica
• Database connectivity
stack
• Self-contained sub-
project of Calcite
• Fast, open, stable
• Powers Phoenix Query
Server

20. Calcite – APIs and SPIs
Cost, statistics
RelOptCost
RelOptCostFactory
RelMetadataProvider
• RelMdColumnUniquensss
• RelMdDistinctRowCount
• RelMdSelectivity
SQL parser
SqlNode
SqlParser
SqlValidator
Transformation rules
RelOptRule
• MergeFilterRule
• PushAggregateThroughUnionRule
• 100+ more
Global transformations
• Unification (materialized view)
• Column trimming
• De-correlation
Relational algebra
RelNode (operator)
• TableScan
• Filter
• Project
• Union
• Aggregate
• …
RelDataType (type)
RexNode (expression)
RelTrait (physical property)
• RelConvention (calling-convention)
• RelCollation (sortedness)
• TBD (bucketedness/distribution)
JDBC driver (Avatica)
Metadata
Schema
Table
Function
• TableFunction
• TableMacro
Lattice

21. Calcite Planning Process
SQL
parse
tree
Planner
RelNode
Graph
Sql-to-Rel Converter
SqlNode
! RelNode
+ RexNode
Node for each node in Input
Plan
Each node is a Set of
alternate Sub Plans
Set further divided into
Subsets: based on traits like
sortedness
1. Plan Graph
Rule: specifies an Operator
sub-graph to match and logic
to generate equivalent ‘better’
sub-graph
New and original sub-graph
both remain in contention
2. Rules
RelNodes have Cost &
Cumulative Cost
3. Cost Model
Used to plug in Schema,
cost formulas
Filter selectivity
Join selectivity
NDV calculations
4. Metadata Providers
Rule Match Queue
Best RelNode Graph
Translate to
runtime
Logical Plan
Based on “Volcano” & “Cascades” papers [G. Graefe]
Add Rule matches to Queue
Apply Rule match transformations
to plan graph
Iterate for fixed iterations or until
cost doesn’t change
Match importance based on cost of
RelNode and height

22. Views and materialized views
• A view is a named
relational expression,
stored in the catalog,
that is expanded
while planning a
query.
• A materialized view is an equivalence,
stored in the catalog, between a table
and a relational expression.
The planner substitutes the table into
queries where it will help, even if the
queries do not reference the
materialized view.

23. Query using a view
Scan [Emps]
Join [$0, $5]
Project [$0, $1, $2, $3]
Filter [age >= 50]
Aggregate [deptno, min(salary)]
Scan [Managers]
Aggregate [manager]
Scan [Emps]
SELECT deptno, min(salary)
FROM Managers
WHERE age >= 50
GROUP BY deptno
CREATE VIEW Managers AS
SELECT *
FROM Emps
WHERE EXISTS (
SELECT *
FROM Emps AS underling
WHERE underling.manager = emp.id)
view scan to
be expanded

24. After view expansion
Scan [Emps] Aggregate [manager]
Join [$0, $5]
Project [$0, $1, $2, $3]
Filter [age >= 50]
Aggregate [deptno, min(salary)]
Scan [Emps]
SELECT deptno, min(salary)
FROM Managers
WHERE age >= 50
GROUP BY deptno
CREATE VIEW Managers AS
SELECT *
FROM Emps
WHERE EXISTS (
SELECT *
FROM Emps AS underling
WHERE underling.manager = emp.id)
can be pushed
down

25. Materialized view
Scan [Emps]
Aggregate [deptno, gender,
COUNT(*), SUM(sal)]
Scan [EmpSummary]
=
Scan [Emps]
Filter [deptno = 10 AND gender = ‘M’]
Aggregate [COUNT(*)]
CREATE MATERIALIZED VIEW EmpSummary AS
SELECT deptno,
gender,
COUNT(*) AS c,
SUM(sal) AS s
FROM Emps
GROUP BY deptno, gender
SELECT COUNT(*)
FROM Emps
WHERE deptno = 10
AND gender = ‘M’

26. Materialized view, step 2: Rewrite query to
match
Scan [Emps]
Aggregate [deptno, gender,
COUNT(*), SUM(sal)]
Scan [EmpSummary]
=
Scan [Emps]
Filter [deptno = 10 AND gender = ‘M’]
Aggregate [deptno, gender,
COUNT(*) AS c, SUM(sal) AS s]
Project [c]
CREATE MATERIALIZED VIEW EmpSummary AS
SELECT deptno,
gender,
COUNT(*) AS c,
SUM(sal) AS s
FROM Emps
GROUP BY deptno, gender
SELECT COUNT(*)
FROM Emps
WHERE deptno = 10
AND gender = ‘M’

27. Materialized view, step 3: Substitute table
Scan [Emps]
Aggregate [deptno, gender,
COUNT(*), SUM(sal)]
Scan [EmpSummary]
=
Filter [deptno = 10 AND gender = ‘M’]
Project [c]
Scan [EmpSummary]
CREATE MATERIALIZED VIEW EmpSummary AS
SELECT deptno,
gender,
COUNT(*) AS c,
SUM(sal) AS s
FROM Emps
GROUP BY deptno, gender
SELECT COUNT(*)
FROM Emps
WHERE deptno = 10
AND gender = ‘M’

28. (and now, back to Phoenix)

29. Example 1, Revisited: Secondary Index
Optimizer internally creates a mapping (query, table) equivalent to:
Scan [Emps]
Filter [deptno BETWEEN 100 and 150]
Project [deptno, name]
Sort [deptno]
CREATE MATERIALIZED VIEW I_Emp_Deptno AS
SELECT deptno, empno, name
FROM Emps
ORDER BY deptno
Scan [Emps]
Project [deptno, empno, name]
Sort [deptno, empno, name]
Filter [deptno BETWEEN 100 and 150]
Project [deptno, name]
Scan
[I_Emp_Deptno]
1,000
1,000
200
1600 1,000
1,000
200
very simple
cost based
on row-count

30. Beyond Phoenix 4.8
with Apache Calcite
• Get the missing SQL support
– WITH, UNNEST, Scalar subquery, etc.
• Materialized views
– To allow other forms of indices (maybe defined as external), e.g., a
filter view, a join view, or an aggregate view.
• Interop with other Calcite adapters
– Already used by Drill, Hive, Kylin, Samza, etc.
– Supports any JDBC source
– Initial version of Drill-Phoenix integration already working

31. Drillix: Interoperability with Apache Drill
SELECT deptno, sum(salary) FROM emps GROUP BY deptno
Stage 1:
Local Partial aggregation
Stage 3:
Final aggregation
Stage 2:
Shuffle partial results
Drill Aggregate [deptno, sum(salary)]
Drill Shuffle [deptno]
Phoenix Aggregate [deptno, sum(salary)]
Phoenix TableScan [emps]
Phoenix Tables on HBase

32. Thank you! Questions?
@maryannxue
@julianhyde
http://phoenix.apache.org
http://calcite.apache.org